| 摘要: |
| 【目的】比较3种标准树高曲线建立方法的优劣，为选择适宜的标准树高曲线建立方法提供依据。【方法】 以福建省将乐县国有林场29块杉木人工林实测数据为依据，采用传统非线性模型、BP神经网络模型、非线性混合模型分别建立杉木标准树高曲线模型，以决定系数R2、均方根误差RMSE以及平均绝对残差E为模型评价和检验指标，对比分析三者的拟合效果。【结果】从拟合精度来看，非线性混合模型、BP神经网络模型、传统模型的决定系数分别为0.916 1，0.904 8和0.889 7，RMSE分别为1.652 9，1.761 2和1.895 4，E分别为1.205 9，1.291 7和1.400 1；从预测精度来看，三者的决定系数分别为0.941 5，0.935 2和0.918 3，RMSE分别为1.361 8，1.432 2和1.609 0，E分别为0.989 8，1.030 5和1.142 8。【结论】3种方法均能较好地模拟杉木树高的生长，BP神经网络模型与非线性混合模型的拟合精度和预测能力均较传统的非线性模型好，但非线性混合模型略优于BP神经网络模型。 |
| 关键词: 杉木 标准树高曲线 传统非线性模型 BP神经网络 非线性混合模型 |
| DOI： |
| 分类号: |
| 基金项目:国家林业局重点项目（2012-07）；林业公益性行业科研专项(200904003-1)；林业科技成果国家级推广项目（［2014］26） |
|
| Comparison of three methods for constructing generalized height-diameter curve |
|
DONG Yun-fei,SUN Yu-jun,XU Hao
|
| Abstract: |
| 【Objective】The advantages of three methods for constructing generalized height-diameter curve were compared to provide basis for choosing optimal method.【Method】The data from 29 Cunninghamia lanceolata plots located in national forest farm of Jiangle in FuJian were used to develop the generalized height diameter curve for C.lanceolata.The curve was built by traditional nonlinear model,BP neural network model and nonlinear mixed model,respectively.The simulation effects of three models were compared using coefficient of determination (R2),root mean square error (RMSE),and absolute mean error (E).【Result】Based on fitting accuracy,the coefficients of determination of nonlinear mixed model,BP neural network model and traditional nonlinear model were 0.916 1,0.904 8,and 0.889 7,the root mean square errors were 1.652 9,1.761 2,and 1.895 4,and the absolute mean errors were 1.205 9,1.291 7,and 1.400 1,respectively.Based on prediction precise,the coefficients of determination were 0.941 5,0.935 2,and 0.918 3,and the root mean square errors were 1.361 8,1.432 2,and 1.609 0,and the absolute mean errors were 0.989 8,1.030 5,and 1.142 8,respectively.【Conclusion】All three methods could simulate height growth of C.lanceolata well.The simulation and prediction accuracies of BP neural network model and nonlinear mixed model were better than that of traditional nonlinear model.Nonlinear mixed model was slightly better than BP neural network model. |
| Key words: Cunninghamia lanceolata generalized height diameter model traditional nonlinear model BP neural network nonlinear mixed model |
